neural-style-tf

Why we included this project

Anyone wanting to reproduce the classic Gatys style-transfer results from the 2016 papers in TensorFlow will find this one of the more complete reference implementations to study. Beyond the basic content-plus-style recipe, it bundles the extras that make the technique usable: blending several style images with controllable interpolation, color-preserving options that keep the original photo's palette, and a route for stylizing video frames with temporal consistency. A segmentation mode even lets you apply one style to the foreground and another to the background of a single scene. The code is readable and lives in single scripts, which makes it a good baseline for researchers and developers to dissect and modify, though it targets the older TensorFlow 1.x APIs rather than a maintained production service.